AI-baserad Måldetektering och Parameterestimering för Bistatisk Radar

Target detection within radar systems is a growing field. Specifically, there is an interest in exploring Artificial Intelligence (AI) models for detecting the presence of targets in bistatic radar data while also estimating target parameters like range and Doppler frequency. This thesis develops two AI-based models. The first is called Range-Doppler estimation Network (RaDeNet) and detects targets and estimates their range and Doppler parameters on a snapshot of bistatic radar data. The second is called Range-Doppler estimation Network-Time (RaDeNet-T) and outputs the same as RaDeNet, however is trained on several time-sequential snapshots of radar data. The performance of these models are compared against the traditional target detection method Ordered Statistics-Constant False Alarm Rate (OS-CFAR) for the target detection task and OS-CFAR with curve fitting for the parameter estimation task. Investigations are made into the performances of these models, together with investigating how varying Signal to Noise (SNR) ratios affected these performances. Results show that a both AI-based models outperformed OS-CFAR in the detection of targets. And RaDeNet-T performed the best in accurately estimating the range and Doppler parameters. Furthermore, RaDeNet-T was also the most affected as SNR levels varied. These results give further support for the potential of AI-based models in target detection and parameter estimation for bistatic radar.